<p>The abstract serves both as a general introduction to the topic and as a brief, non-technical summary of the main results and their implications. Authors are advised to check the author instructions for the journal they are submitting to for word limits and if structural elements like subheadings, citations, or equations are permitted. At present, there are some conditions in the road environment, such as large changes in vehicle target size, small target dense occlusion, high overlap rate, and insufficient lighting. In this complex case, the existing detection model has the problem that the target detection results are not accurate. To address these problems, this paper proposes a BML-YOLO, a multiscale object detection method based on feature fusion. First, to strengthen the feature extraction, this paper uses the idea of a bidirectional feature pyramid and an adaptive spatial feature fusion network in the neck network. It proposes a new feature fusion network (BAFPN). Secondly, to speed up the model convergence and improve the regression accuracy of the prediction box, a new penalty term is added to the loss function to propose a new loss function (MCIOU). The loss function introduces the minimum diagonal ratio of the rectangular box as the penalty term to compare the similarity of the bounding box and further improve the detection performance. Finally, in order to improve the extraction ability of spatial feature information and strengthen the generalization ability of the model, the ordinary convolution and depth separable convolution combination module (LCBS) is constructed to further improve the detection performance. The datasets used for the experiment are the KITTI dataset and the vehicle STREETS dataset. The results show that the proposed algorithm is improved by 5.5<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4328_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> and 4.1<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4328_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> for mAP@0.5 on the KITTI and STREET datasets, and the mAP@0.5:0.9 increased by 9.6<InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4328_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> and 2.4<InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4328_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation>, respectively. The method is better than the traditional target detection network in other key indicators, such as recall rate and accuracy.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

BML-YOLO: Multi-scale vehicle target detection method based on feature fusion

  • Xinying Chen,
  • Ying Liu,
  • Shuyuan Li

摘要

The abstract serves both as a general introduction to the topic and as a brief, non-technical summary of the main results and their implications. Authors are advised to check the author instructions for the journal they are submitting to for word limits and if structural elements like subheadings, citations, or equations are permitted. At present, there are some conditions in the road environment, such as large changes in vehicle target size, small target dense occlusion, high overlap rate, and insufficient lighting. In this complex case, the existing detection model has the problem that the target detection results are not accurate. To address these problems, this paper proposes a BML-YOLO, a multiscale object detection method based on feature fusion. First, to strengthen the feature extraction, this paper uses the idea of a bidirectional feature pyramid and an adaptive spatial feature fusion network in the neck network. It proposes a new feature fusion network (BAFPN). Secondly, to speed up the model convergence and improve the regression accuracy of the prediction box, a new penalty term is added to the loss function to propose a new loss function (MCIOU). The loss function introduces the minimum diagonal ratio of the rectangular box as the penalty term to compare the similarity of the bounding box and further improve the detection performance. Finally, in order to improve the extraction ability of spatial feature information and strengthen the generalization ability of the model, the ordinary convolution and depth separable convolution combination module (LCBS) is constructed to further improve the detection performance. The datasets used for the experiment are the KITTI dataset and the vehicle STREETS dataset. The results show that the proposed algorithm is improved by 5.5 \(\%\) % and 4.1 \(\%\) % for mAP@0.5 on the KITTI and STREET datasets, and the mAP@0.5:0.9 increased by 9.6 \(\%\) % and 2.4 \(\%\) % , respectively. The method is better than the traditional target detection network in other key indicators, such as recall rate and accuracy.